Universal Cart, UCP, AP2: what agentic commerce changes for e-commerce brands

Agentic commerce scales up

On May 19, 2026, at Google I/O, Google announced Universal Cart, a cart that works across Search, the Gemini app, YouTube and Gmail. It aggregates items added from any Google surface, tracks prices in the background, anticipates incompatibilities, automatically applies Google Wallet loyalty programs, and offers integrated checkout via Google Pay or a handoff to the merchant site.

The announcement fits a clear trajectory. The Universal Commerce Protocol was published earlier this year, the Agentic Commerce Protocol (AP2) secures payments made by an agent, and Universal Cart becomes the visible consumer endpoint.

For e-commerce teams, the question is no longer whether agents will buy on behalf of users. The question is: is your catalog ready to be read, understood, compared and recommended by an agent?

What concretely changes for a retailer

Three things move at the same time.

The decision point shifts from the PDP to the cart. Previously, the purchase decision was made on the merchant’s product page. Universal Cart introduces an intermediate layer that compares, flags price drops, checks compatibility between components, and applies loyalty cards. The merchant loses part of the control over the moment of decision.

Checkout becomes triggerable from Google. Google Pay now enables checkout in a few clicks at Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and at selected Shopify merchants like Fenty and Steve Madden. The merchant still owns the transaction, but orchestration moves out of its domain.

The agent takes a growing share of purchase responsibility. With AP2, the user defines authorized brands, products and budgets, and the agent checks out only if those criteria are met. The retailer dialogues with a software buyer, not a human one. The signals a human sends (hesitation, scrolling, going back, abandoning the cart) disappear.

Universal Cart, UCP, AP2: how the layers fit together

Many teams still confuse these three names. They do not operate at the same level.

UCP (Universal Commerce Protocol) is the data and action standardization layer. It defines how an agent describes a product, a cart, a promo, a stock state. Without UCP, every merchant speaks a different dialect. With UCP, agents read catalogs the same way across compatible sites. Our complete UCP guide details the Merchant Center attributes that follow from it.

AP2 (Agentic Commerce Protocol) is the secure payment layer. It ensures a verifiable link between user, merchant and payment processor, with tamper-proof digital authorizations that create a permanent trace. In case of return, merchant and buyer share the same proof.

Universal Cart is the consumer experience that assembles both. It is the visible interface. UCP and AP2 are the underlying layers.

From a generative search perspective, this articulation has a direct consequence. An agent that wants to recommend a product must read UCP attributes, verify availability, and know the payment will succeed. If any one of those three is missing, the product is silently dropped.

Signals an agent reads, ignores, or misinterprets

This section is not theoretical. It reflects what we observe across recent deployments.

Reliably read signals

• Properly formed Schema.org Product and Offer with gtin, price, priceCurrency, availability, itemCondition.
• Filled Merchant Center attributes: gtin, brand, mpn, product_type, google_product_category, availability_date.
• FAQs structured in JSON-LD with a recent last-update.
• Verified reviews exposed with aggregateRating and reviewCount.
• Clear titles and descriptions, without keyword stuffing.

Regularly misinterpreted signals

• Variants (size, color, format) when attributes are not explicit. The agent may recommend the wrong variant.
• Real-time availability: a stock marked as available but actually unavailable breaks the agent’s trust.
• Compatibilities. A USB-C cable compatible only with certain models, without an explicit attribute, will be excluded from configuration recommendations.

Ignored signals

• Content rendered only in client-side JavaScript without server rendering. OpenAI and Gemini agents read part of it, but not reliably.
• Text hidden in carousels loaded on demand.
• Information buried in PDFs or undescribed images.

The Dataiads Agentic Commerce Index audits these three layers in 30 seconds on a given PDP and produces a readiness score for ChatGPT and Gemini.

Where it breaks in production

At the scale of a large catalog, certain failure patterns recur.

Wide, heterogeneous catalogs. On catalogs of several hundred thousand SKUs, attribute quality varies by category, brand, supplier. Secondary product pages (rare variants, imported products, third-party brands) are under-equipped in metadata and invisible to an agent.

Stock synchronization. Universal Cart tracks prices and availability in the background. A PDP showing “in stock” while the feed says “out of stock” will be flagged as unreliable on the agent side. Five minutes of lag is acceptable. Several hours is not.

Multi-market. The same product sold in France, Italy and Spain with different prices, sizes and labels requires consistent local feeds. Agents compare language versions. A major inconsistency between two markets reduces the probability of recommendation.

Category pages used as PDPs. The agent looks for a product page. A page grouping 30 references with no clear canonical will not be extracted as a product.

Fear, excitement, adoption: where retailers stand today

The ecosystem experiences three feelings in parallel.

Fear comes from loss of control. If the agent decides which product is recommended, the merchant worries about a concentration of sales on a few “safe” references, at the expense of the long tail. Challenger brands fear a reinforced Matthew effect.

Excitement comes from brands that already invested in feed quality. For them, it is a new acquisition channel with a known entry cost and high transactional intent. Sephora, Target, Walmart and Wayfair position first in Universal Cart for operational reasons, not marketing ones: their catalogs are already structured. The Sephora in ChatGPT case illustrates the cohabitation model between agentic discovery and merchant checkout.

Adoption happens in two waves. Merchant Center-compliant retailers benefit from the Universal Cart rollout in the United States this summer, followed by Canada, Australia and the United Kingdom. The others watch their UCP conformance rate climb quarter after quarter.

What your teams should do before Universal Cart reaches Europe

Not a step-by-step guide. A list of conditions to validate.

1. Audit how your PDPs are read by an agent. Load a PDP, check that the raw HTML contains the name, price, availability and critical attributes without relying on JavaScript rendering. The full methodology is documented in the Agentic Commerce Index methodology.
2. Verify UCP conformance on your Merchant Center feed. Required and recommended attributes have been broadened for agentic commerce. Our article on Merchant Center attributes in the UCP era lists the concrete changes.
3. Update stock synchronization frequency. Move from daily to hourly or real-time sync for high-rotation SKUs. The gap between feed and stock reality is an eviction signal on the agent side.
4. Structure product FAQs in JSON-LD. Agents extract FAQ answers during the semantic decisioning phase. An unstructured FAQ does not contribute to recommendations.
5. Audit variants. Each variant must have its own identifier, GTIN, and distinctive attributes. An implicit variant is an invisible variant.
6. Plan the AP2 transition. The agentic payment protocol requires cryptographic authorizations on the merchant side. Google starts integrating AP2 into its products over the coming months, starting with Gemini Spark. The retailer infrastructure must follow.

The game is no longer to rank, it is to be citable

The mental model that drove SEO and paid media for fifteen years assumes a human clicks on a result. Universal Cart breaks that model. The user no longer lands on the PDP: the agent already visited it.

This does not mean traffic disappears. It means a growing share of purchase decisions is made in a zone you do not see in Google Analytics, but that is conditioned by the structural quality of your catalog. LLMO and GEO are not disciplines parallel to SEO: they are the conditions of visibility in that zone. Our LLMO for e-commerce guide  covers the optimization mechanics for large language models.

For paid media teams, the shift is just as sharp. Performance Max keeps working, but the battle angle moves: bids and creatives are not the only performance lever anymore. Product feed readability upstream is. A PMax campaign on a poorly attributed product hits a structural ROAS ceiling. We documented this in the AI and paid media analysis.

Key takeaways

• Universal Cart is the consumer interface of Google’s agentic ecosystem. UCP is the data standardization layer, AP2 is the payment layer.
• The purchase decision shifts from the PDP to the agentic cart. The retailer loses part of the control over the conversion moment.
• Product feed and PDP readability by an agent becomes a standalone performance factor, measurable independently from traditional SEO.
• Highest-impact signals: Schema.org Product/Offer, UCP attributes, stock synchronization, structured FAQ, explicit variant handling.
• Merchant Center-compliant retailers are in the first deployment wave. Others have a few quarters to close the gap.
• Universal Cart rolls out in the United States this summer, followed by Canada, Australia and the United Kingdom. France will follow.

FAQ

What is Universal Cart?

Universal Cart is a universal cart announced by Google at I/O 2026. It works across Search, the Gemini app, YouTube and Gmail, aggregates items from each of these surfaces, tracks prices, anticipates incompatibilities, and offers integrated checkout via Google Pay or a handoff to the merchant site.

What is the difference between UCP and AP2?

UCP (Universal Commerce Protocol) standardizes how agents read catalogs: product attributes, price, stock, promos. AP2 (Agentic Commerce Protocol) standardizes how agents pay on your behalf, with cryptographic guardrails and a verifiable trace. UCP is the data layer, AP2 is the transaction layer.

Is Universal Cart available in France?

Not yet. Google announced a United States rollout in summer 2026, followed by Canada, Australia and the United Kingdom later in the year. France has no official date, but UCP and AP2 are already public and adopted by several technology partners.

What does a retailer need to do to be visible in Universal Cart?

Three priorities: Merchant Center feed compliance with UCP attributes, PDP HTML readability by an agent without JavaScript rendering, fast stock synchronization. The Dataiads Agentic Commerce Index audits a PDP in 30 seconds and identifies critical gaps.

What is the impact on Performance Max?

PMax keeps working, but the performance ceiling is now more directly tied to product feed quality than before. A poorly attributed product hits a ROAS ceiling, regardless of bids and creatives. Agentic readability becomes an upstream performance factor.

How does Universal Cart change the role of merchant checkout?

The merchant remains the owner of the transaction and the customer data. But the decision moment moves upstream, into the agentic cart. This forces the merchant to send price, stock, promotion and loyalty signals directly consumable by Google, rather than relying on the visit and navigation on the site.

‍

‍